Polycentrism in cities is studied with the help of an agent-based model grounded in the Alonso, Muth, Mills (AMM) framework, using microeconomic interactions between heterogeneous agents. This model is shown to reproduce the standard urban equilibrium with two income groups. Two job centers at various distances from each other are introduced, and the economic, social and environmental outcomes of these various polycentric spatial structures are presented. Then we introduce two-worker households whose partners may work in different job centers. If all households are "split", i.e. the two partners work in different job centers, polycentrism is shown to be undesirable. When various two-worker households are mixed we find that polycentrism is desirable, as long as centers are not moved too far apart from each other. The environmental outcome is also positive for small values of this distance, but this positive effect is mitigated by the fact that housing surfaces increase.